then 300 fs and close to the jitter of the 80 Gb/s transmitter /8/.This indicates an excellent system performance of the high speedall-optical clock recovery based on the self-pulsating laser.

5. SummaryOptical clock recovery is a key function for signal processing infuture high speed and flexible all-optical networks. Self-pulsatingDFB lasers are developed for these applications. They are compactsemiconductor devices, easy to operate, and tuneable in frequencyvia the driving dc currents. Their good performance as opticalclock has been demonstrated in several system experiments.Important advantages are the ultra high speed potential exceeding

that of electronics and the fast locking function, needed for opera-tion in asynchronous packet switched networks.

6. References 1. B. Sartorius, OFC 2001, invited paper MG7, Anaheim, Cal, USA 2. M. Radziunas et al, IEEE J. QE 36, pp.1026, 20003. M. Möhrle et al, IEEE J. Sel. Topics in QE, 7, pp. 217, 20014. C. Bornholdt, et al, Electron. Lett. 36, pp.327, 2000.5. B. Sartorius et al, OFC 2000, paper PD 11, Baltimore, USA6. S. Bauer, et al., OFC 2000, paper TuF5, Baltimore, USA7. C. Bornholdt et al, ECOC 2001, paper Th.F.1.2, Amsterdam, NL8. C. Bornholdt et al, OFC 2002, paper TuN, Anaheim, Cal, USA

OCTOBER 2002 IEEE LEOS NEWSLETTER 19

System Perspective for All-Optical Switching

C. BINTJAS, N. PLEROS and H. AVRAMOPOULOS

Department of Electrical and Computer EngineeringNational Technical University of Athens, 15773 Zographou, Athens, Greece

It is true to say that the past year has been one of significantchanges in the telecommunications industry as a result of therecent market downturn. It is also true to say that before this,

the telecommunications industry had experienced an explosivegrowth for a good number of years. For wired networks this growthwas due to the unprecedented improvement in performance of thephotonic technologies as well as proof of their maturity that hasresulted in a spectacular decrease in transmission cost, which inturn has lead to the widespread adoption and construction of newfiber networks. At this point of the business cycle and in prepara-tion for the next crest of the wave, it is appropriate to reviewresearch lines that have been maturing over the years and whoseresults may be incorporated in next generation products. The pur-pose of this article is to argue that all-optical logic and all-opticalswitching techniques are such research themes.

All-optical switching, signal processing and more generally opti-cal computing has been the holy grail of researchers ever since theinvention of the laser. General purpose all-optical processing is stilla long way off and may never materialize as electronics possess andseem capable to maintain huge technical and cost advantages.However there are specialized applications in high data ratetelecommunications, where low complexity all-optical processingcircuits are ideally suitable to provide functionalities that electronicsolutions cannot, so that they may be of commercial value.

In order for all-optical switching techniques to become a seriousconsideration, they must simultaneously possess four properties: (a) aconsiderable speed advantage and otherwise ability to simplify thecircuit design, (b) switching energies that are similar to those of elec-tronics (c) capability for integration and of course (d) an applicationdomain where these advantages may be of use. It was relatively earlyrecognized that for the implementation of all-optical gates, interfero-metric arrangements offered speed advantages and Boolean logiccapability [1,2]. Figure 1 shows a generic interferometric arrange-ment drawn as a Mach-Zehnder interferometer, but which could infact be any type of interferometer as the Sagnac or the Michelson.

The interferometer consists of two separate optical paths where thephase of the optical field may be independently controlled in a non-linear optical medium by optical means. In the example of figure 1,assuming that on each of the two interferometer paths the phase ofthe optical field CLK may be changed by π depending on whetherone or both of the external controlling signals A and B are present,interference on the output coupler causes the result of the Booleanaddition XOR between A and B to be written on CLK. The rest ofthe Boolean operations may be implemented similarly.

Earlier research efforts used optical fiber as the nonlinear materi-al partly due to its ultrafast Kerr nonlinearity and mainly because itwas so much more easily available than for example optical semi-conductor devices. Very spectacular results in terms of speed havebeen achieved using the nonlinear optical fiber, Sagnac interferome-ter, including the demultiplexing of a 10 Gbps channel from a 640Gbps data stream in a 1.28 Tbps transmission experiment [3].Unfortunately the Kerr nonlinearity of optical fiber is weak, requir-ing the use of long pieces of fiber, making the devices hard to inte-grate and their switching energies high. In the example given abovethe interferometer used 450 m of fiber and required about 2 pJ ofswitching energy for 1 ps pulses [4], so that operation of the switch

Figure 1. Generic interferometric gate arrangement configured to perform BooleanXOR between signals A and B. The output is written on signal CLK.

20 IEEE LEOS NEWSLETTER OCTOBER 2002

at the full data rate becomes energy-wise expensive. It was the realization that fast phase variations can also be

obtained in semiconductor optical amplifiers (SOAs) as a result of therelation between the refractive index and the carrier density, that hasmore recently lead to a whole new class of compact and very lowswitching energy devices. These can be broadly separated to eitherdiscrete SOA interferometric gates or integrated devices. The discretedevices are single optical path designs that are either of the dual armSagnac type [5-7], or single arm type [8,9] and include a single SOA.Integrated devices have been demonstrated by both monolithic inte-gration or hybrid integration of SOAs on Si-based planar lightwavecircuits and follow the Michelson or the Mach-Zehnder designs. Thetechnology status of semiconductor optical gates including integra-tion activities have been very thoroughly reviewed in reference [10].The basic nonlinearity that these devices exploit is resonant and forthis reason switching has been shown repeatedly to be possible withenergies of few fJ for pulses of few ps duration [11-13], in very com-pact devices. In this context demultiplexing from 336 Gbps to 10.5Gbps and wavelength conversion experiments have been shown [14],as well as more taxing experiments for the switching element at thefull data rate of 100 Gbps [15].

From the previous discussion it must be obvious that the recentadvances of all-optical switching techniques have addressed the issuesof speed, switching energy and capability for integration. It remainsto clearly identify application regimes for these devices where theydisplay superior suitability compared to alternative solutions.Applications related to ultra-high data rate single channel TDMtransmission such as demultiplexing were identified early. All-opticalregeneration was also identified as a key functionality that can beprovided by optical gates, by taking advantage of their amplitudeand timing jitter suppression properties [16-18]. It was however theoverwhelming penetration of WDM systems and the need for fastwavelength converters that has provided one of the main applicationareas [12, 19-20], because of the very steep transfer function and highextinction ratio that semiconductor interferometric devices can give.

Looking further ahead and in view of achieving better resource

use at the network level, some form ofpacket or burst switching is likely to beneeded. Currently only circuit opticalswitches are available in the market. All-optical packet switches exist only in labo-ratories and the demonstrations availabledo not offer the distinguishing features ofpacket switching, that is, on-demand useof bandwidth and the ability to switchbusty traffic without excessive packet loss.Proposals of how to build ‘true’, opticalpacket switches have been made in the past[21] and these rely on algorithms thatguarantee lossless routing through theswitch and optimal cost scaling of theswitch in terms of the burstiness of thedata. Such a switch consists of three mainunits: (a) the electronic control unit that isresponsible for running the routing algo-rithm from information obtained from thepacket headers, (b) a packet slot inter-changer unit at the input of the switchthat plays the role of buffer to avoid inter-

nal collisions and (c) a space switch for output routing. The routingalgorithm is still performed in the electronic domain, but the con-trol of the optical switching elements can be done all-optically, soas to avoid the complex electronic driver circuits for the opticalswitches and to improve the overall switch speed. To implementsuch design based on the all-optical control of the switching ele-ments, several functionalities are needed including packet synchro-nization [22] and parity checking [23]. Key optical circuits are alsoneeded as, for example, 2x2 exchange bypass switches [13] for thepacket slot interchanger or clock recovery units that may operate ona packet basis. Figure 2 shows a recently demonstrated all-opticalcircuit to acquire clock from packets [24]. The packet clock recov-ery unit uses a Fabry-Perot (FP) filter with a free spectral rangeequal to the data rate followed by a nonlinear, high speed, opticalgate. The role of the FP filter is to partially fill the ‘0s’ of theincoming packet and the optical gate employs its nonlinear transferfunction to equalize the amplitudes of the partially filled ‘1s’.Important features of this circuit are that it acquires clock withinvery few bits, the clock signal persists approximately for the dura-tion of the packet, it does not contain any high speed electronicsand does not require any external synchronization, so that it couldbe suitable to power an all-optically controlled packet switch.

In summary all-optical switching techniques have matured tech-nically, very significantly over the past few years and across allfronts. Advances have spanned from the material and device level,to the system and application level as well as performance and prac-ticality. Improvements were the result of effort in a number of labo-ratories across the world and these have brought all-optical switch-ing technology to the point of commercial exploitation. The nextfew years will call the commercial verdict on these efforts.

References1. A. Lattes, et. al., “An ultrafast all-optical gate,” IEEE J. Quantum

Electron., Vol. 19, pp. 1718-1723, 1983.2. G. Eichmann, Y. Li and R. R. Alfano, “Digital optical logic using a

Fab

ry-P

erot

Filt

er

localcw laser

Opt

ical

Gat

e

packetspackets packet clockpacket clock

Figure 2. All-optical clock recovery from short data packets at 10 Gbps. Note the short rise time of the packet clock andthat its duration is approximately equal to the original packet length.

OCTOBER 2002 IEEE LEOS NEWSLETTER 21

pulsed Sagnac interferometer switch,” Opt. Eng., vol. 25, p. 91, 1986.3. M. Nakazawa, T. Yamamoto and K. R. Tamura, “1.28 Tbit/s-70

km OTDM transmission using third- and fourth-order simultane-ous dispersion compensation with a phase modulator,” Electron.Lett. , Vol. 36, pp. 2027-2029, 2000.

4. T. Yamamoto, E. Yoshida and M. Nakazawa, “Ultrafast nonlinearoptical loop mirror for demultiplexing 640 Gbit/s TDM signals,”Electron. Lett., Vol. 34, 1013-1014, 1998.

5. J. P. Sokoloff, et. al., “A terahertz optical asymmetric demultiplexer(TOAD),” IEEE Photon. Technol. Lett., Vol. 5, pp. 787-790, 1993.

6. A. D. Ellis, et. al., “Ultra-high-speed OTDM networks using semi-conductor amplifier-based processing nodes,” J. of LightwaveTechnol., Vol. 13, pp. 761-770, 1995.

7. M. Eiselt, W. Pieper and H. G. Weber, “SLALOM: semiconductorlaser amplifier in a loop mirror,” J. of Lightwave Technol, Vol. 13,pp. 2099-2112, 1995.

8. K. Tajima, S. Nakamura, Y. Sugimoto, “Ultrafast polarization-dis-criminating Mach-Zehnder all-optical switch,” Appl. Phys. Lett.,Vol. 67, pp. 3709-3711, 1995.

9. N. S. Patel, K. A. Rauschenbach and K. L. Hall, “40-Gb/s demulti-plexing using an ultrafast nonlinear interferometer (UNI),” IEEEPhoton. Technol. Lett., Vol. 8, pp. 1695-1697, 1996.

10.K. E. Stubkjaer, “Semiconductor optical amplifier-based all-opticalgates for high-speed optical processing,” IEEE J. Select. TopicsQuant. Electron., 6, 1428-1435, 2000.

11.C. Bintjas, et. al, “20 Gb/s all-optical XOR with UNI gate, IEEEPhoton. Technol. Lett., Vol. 12, pp. 834-836, 2000.

12. Y. Ueno, et. al., “168-Gb/s OTDM wavelength conversion using anSMZ-type all-optical switch,” in Proc. ECOC 2000, pp. 13-14, 2000.

13.G. Theophilopoulos, et. al., “Optically addressable 2x2 exchange

bypass packet switch,” IEEE Photon. Technol. Lett., Vol. 14, pp.998-1000, 2002.

14. S. Nakamura, Y. Ueno and K. Tajima, “Ultrahigh-speed opticalsignal processing with Symmetric-Mach-Zehnder-type all-opticalswitches,” TuK4 in Technical Digest 2002 LEOS Summer TopicalMeetings, Mont Tremblant, Quebec, Canada, 2002.

15.K. L. Hall and K. A. Rauschenbach, “100-Gbit/s bitwise logic,”Opt. Lett., Vol. 23, pp. 1271-1273, 1998.

16.H. Avramopoulos and N. A Whitaker Jr., “Optical regenerationcircuit,” US patent No. 5369520, 1994.

17.B. Lavigne, et. al., “Full validation of an optical 3R regenerator at20 Gbit/s”, in Proc. OFC 2000, Vol. 3, pp. 93-95, 2000.

18.B. Sartorius, “3R All-optical signal regeneration”, in Proc. ECOC2001, Vol. 5, pp. 98-125, 2001.

19.C. Joergensen, et. al., “Wavelength conversion by optimized mono-lithic integrated Mach-Zehnder interferometer,” IEEE Photon.Technol. Lett., Vol. 8, pp. 521-523, 1996.

20.J. Leuthold, et. al., “All-optical wavelength conversion up to 100Gb/s with SOA delayed-interference configuration,” in Proc. ECOC2000, pp. 119-120, 2000.

21. E. A. Varvarigos, “The ‘Packing’ and ‘Scheduling packet’ switcharchitectures for almost all-optical lossless networks,” J. ofLightwave Technol. Vol. 16, pp. 1757-1767, 1998.

22.S. A Hamilton and B. S. Robinson, “40-Gb/s all-optical packet syn-chronization and address comparison for OTDM networks,” IEEEPhoton. Technol. Lett., Vol. 14, pp. 209-211, 2002.

23.J. Poustie, et. al., “All-optical parity checker with bit-differentialdelay,” Opt. Commun., Vol. 162, pp. 37-43, 1999.

24.C. Bintjas, et. al., “Clock Recovery Circuit for Optical Packets,” tobe published in IEEE Photon. Technol. Lett., September 2002.

Fast Processes in Semiconductor Optical Amplifiers: Theory and Experiment

JESPER MØRK, TOMMY W. BERG and SVEND BISCHOFF

COM, Technical University of Denmark, Building 345V, DK-2800 Kgs. Lyngby, DenmarkE-mail: jm@com.dtu.dk, Tel: +45 4525 5765, Fax: +45 4593 6581

AbstractWe review the physical processes that are responsible for ultrafastgain and index dynamics in semiconductor optical amplifiers andwhich impact high-speed optical switching applications.

All-optical signal processing is expected to play an important rolein future high-capacity optical communication systems. Primaryincentives for this evolution are the larger data rates that can be han-dled by all-optical devices, as well as the cost-reduction and increasedflexibility that may be achieved by avoiding conversions between theoptical and the electronic domain. In order to be practical and com-petitive, all-optical switching devices should fulfil criteria similar tothose of electronics, i.e., be small and allow integration of differentfunctionalities, and have the potential for cheap mass-production.Presently, semiconductor optical amplifier (SOA) based devices areamong the primary contenders for integrated all-optical devices. The

large gain and large differential gain of SOAs allow switching withpower levels in the range of milliwats, and various functionalitieshave been demonstrated in a number of different schemes at speeds inexcess of 100 Gb/s, e.g. wavelength conversion at 168 Gb/s [1]. Inthis paper we will review and discuss the physical processes thatimpact the operation of SOAs at such high data-rates.

Common to the various schemes utilizing SOAs as the mainswitching element is the exploitation of saturation effects in theactive region of the waveguide. When an optical beam is injected inthe amplifier, the gain of the amplifier is saturated and, in conse-quence of the induced change of the carrier density in the activeregion, the refractive index of the waveguide is changed, cf. thequalitative illustration in Fig. 1. Both effects can be utilized for all-optical switching of a data signal [2]. In the simplest scheme ofcross-gain modulation (XGM), the gain change simply controls theamplitude of another (probe) beam transmitted through the wave-